diagnose

Raise the reproduction rate instead of chasing a clean repro: loop the trigger 100 times, parallelise, add stress, narrow timing windows, or inject sleeps. A 50%-flake bug is debuggable; keep increasing the rate until hypothesis testing becomes practical.

Use the included human-in-the-loop bash template (scripts/hitl-loop.template.sh). It prompts the user through structured steps, captures their answers as KEY=VALUE output, and feeds the results back to the agent so the loop stays structured.

Build a deterministic feedback loop, reproduce the exact user-reported symptom, and write 3–5 ranked falsifiable hypotheses with explicit predictions. Then write the regression test before the fix, but only if a correct test seam exists.

Establish a baseline measurement first using a timing harness, profiler, or query plan, then bisect against it. Logs are usually the wrong tool for performance work; measure first and fix second.

Stop and say so explicitly, listing what you tried. Ask the user for environment access, a captured artifact such as a HAR file or core dump, or permission to add temporary production instrumentation rather than hypothesising blindly.

Skip it when no correct seam exists — for example when a unit test cannot replicate the multi-caller chain that triggered the bug. A shallow test gives false confidence; document the missing seam as an architectural finding instead.